Social Vulnerability Analysis in Central Java With K-Medoids Algorithm

To address the limitations of the Social Vulnerability Index (SoVI) in only providing a general overview without pinpointing areas of social vulnerability, a correlational approach paired with a clustering method can be applied. This approach helps in identifying dominant factors and pinpointing socially vulnerable districts or cities in Central Java. The study employs the K-Medoids algorithm, which is advantageous when dealing with outliers in the dataset. Three different distance measures are considered: Euclidean, Manhattan, and Minkowski distances, to identify the optimal clustering of social vulnerability. The evaluation of the best cluster is conducted using the Davies-Bouldin Index, a metric for validating clustering models by averaging the similarity of each cluster to its most similar counterpart. Findings indicate that using the K-Medoids algorithm with Manhattan distance yields the most effective clustering, resulting in two distinct clusters. Cluster 1, comprising 25 districts/cities, is identified as the most vulnerable to natural disasters and challenges in education, demography, economy, and health. Meanwhile, Cluster 2, encompassing 10 districts/cities, includes urban areas with the highest social vulnerability, notably in the proportion of rental housing.